import os
import gradio as gr
import torch
import spaces
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load model and tokenizer if a GPU is available
if torch.cuda.is_available():
    model_id = "allenai/OLMo-7B-Instruct"
    model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True)
    tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
else:
    raise EnvironmentError("CUDA device not available. Please run on a GPU-enabled environment.")

# Basic function to generate response based on passage and question
@spaces.GPU
def generate_response(passage: str, question: str) -> str:
    # Prepare the input text by combining the passage and question
    messages = [{"role": "user", "content": f"Passage: {passage}\nQuestion: {question}"}]
    inputs = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(model.device)
    
    # Generate text, focusing only on the new tokens added by the model
    outputs = model.generate(**inputs, max_new_tokens=150)
    
    # Decode only the generated part, skipping the prompt input
    # generated_tokens = outputs[0][inputs.input_ids.shape[-1]:]  # Ignore input tokens in the output
    response = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
    
    return response


# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("# Passage and Question Response Generator")
    
    passage_input = gr.Textbox(label="Passage", placeholder="Enter the passage here", lines=5)
    question_input = gr.Textbox(label="Question", placeholder="Enter the question here", lines=2)
    
    output_box = gr.Textbox(label="Response", placeholder="Model's response will appear here")
    
    submit_button = gr.Button("Generate Response")
    submit_button.click(fn=generate_response, inputs=[passage_input, question_input], outputs=output_box)

# Run the app
if __name__ == "__main__":
    demo.launch()